docs(data-designer-retrieval-sdg): use NVIDIA-NeMo plugin index in README install steps (#47)

The plugin is distributed from the NVIDIA-NeMo plugin index on GitHub
Pages, not PyPI. The README still showed a bare
`pip install data-designer-retrieval-sdg` instruction, which 404s.

Replace the Installation section with three install variants, all
layering the NVIDIA-NeMo index alongside PyPI so transitive dependencies
(data-designer, nltk, pyarrow, pyyaml) still resolve from PyPI:

- `uv pip install` for one-shot installs
- `uv add` for projects managed by uv
- `pip install` with `--index-url` / `--extra-index-url` for pip users

Also drop the stray "single PyPI package" / "the same pip install"
phrasings in the Plugins section.

Verified end-to-end in a fresh venv: install resolves
`data-designer-retrieval-sdg==0.1.0` from the NVIDIA-NeMo index, the
CLI binary works, and both `embedding-dedup` and `document-chunker`
entry points pass `assert_valid_plugin()`.

Signed-off-by: Steve Han <sthan@nvidia.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

Author: shan-nvidia

plugins/data-designer-retrieval-sdg/README.md

@@ -9,16 +9,16 @@ finetuning.
 
 ## Plugins
 
-The single PyPI package contributes two plugins to DataDesigner's
-registries via `[project.entry-points."data_designer.plugins"]`:
+A single package contributes two plugins to DataDesigner's registries
+via `[project.entry-points."data_designer.plugins"]`:
 
 | Slug | Type | Purpose |
 |------|------|---------|
 | `embedding-dedup` | column generator | Generic cosine-similarity dedup of any list-valued column. Implements native `agenerate()` for the async engine. |
 | `document-chunker` | seed reader | Sentence-chunks a directory of text files and emits structured sections, with optional multi-document bundling. |
 
-Both ship with the same `pip install data-designer-retrieval-sdg` and
-become discoverable automatically through Python entry points.
+Both are registered automatically through Python entry points when the
+package is installed (see [Installation](#installation)).
 
 ## Native async (`DATA_DESIGNER_ASYNC_ENGINE=1`)
 
@@ -36,10 +36,40 @@ runs on Python 3.10+ via the framework's sync bridge.
 
 ## Installation
 
+The package is distributed from the NVIDIA-NeMo plugin index (hosted on
+GitHub Pages); it is not on PyPI. Install it by adding the plugin index
+alongside PyPI:
+
+```bash
+uv pip install \
+  --default-index https://pypi.org/simple/ \
+  --index https://nvidia-nemo.github.io/DataDesignerPlugins/simple/ \
+  data-designer-retrieval-sdg
+```
+
+For projects managed with `uv`, add it as a dependency:
+
 ```bash
-pip install data-designer-retrieval-sdg
+uv add \
+  --default-index https://pypi.org/simple/ \
+  --index https://nvidia-nemo.github.io/DataDesignerPlugins/simple/ \
+  data-designer-retrieval-sdg
 ```
 
+`pip` users can pass the equivalent flags:
+
+```bash
+pip install \
+  --index-url https://pypi.org/simple/ \
+  --extra-index-url https://nvidia-nemo.github.io/DataDesignerPlugins/simple/ \
+  data-designer-retrieval-sdg
+```
+
+Standard version constraints work (`>=0.1`, `==0.1.0`, ...). The
+NVIDIA-NeMo index only serves `data-designer-*` plugin packages; the
+default PyPI index supplies transitive dependencies (`data-designer`,
+`nltk`, `pyarrow`, `pyyaml`).
+
 For development inside the monorepo:
 
 ```bash